osFoundry / Platform
osIndex
Search every local file with three scorers — lexical, semantic, recency. Stays on-device.
What it is
osIndex is osFoundry's local file search. Point it at a folder and it makes everything inside findable by three complementary scorers at once: a lexical scorer for exact keywords and identifiers, a semantic scorer that matches on meaning using vector embeddings, and a recency scorer that favors what you touched most recently. A pure keyword search misses files where you remember the concept but not the exact phrase; a pure semantic search misses precise symbol names. Running all three and fusing their rankings is what lets "handle auth" reach both a literal AuthToken and a function called checkSession, with the file you edited this morning weighted up.\n\nInside osFoundry, osIndex is built on-device. Selected folders are indexed into local SQLite with a built-in vector store for the semantic dimension, and the three ranked lists are combined with reciprocal-rank fusion rather than one method overriding the others. Indexing is incremental: a folder watcher re-indexes only the files that changed, so the index stays fresh without a full rescan on every keystroke. Because osFoundry is local-first, your folder contents do not have to leave the machine to become searchable — the index, the vectors, and the query all live on your device.\n\nThis matters because search is only half the point; the other half is grounding. osIndex results become context for your chats and agents, so when you ask a question about your own project, the model answers from your actual files instead of guessing. And since osFoundry is bring-your-own-key, the model calls that read those results run on your own API keys with no per-seat fees and no reseller markup — you keep the retrieval private and pay providers directly for the reasoning.
Key capabilities
- Search across all your local files with three scorers working together: exact keyword (lexical), meaning-based (semantic vectors), and recency, so a query hits the right file whether you remember its wording or just its idea.
- Runs fully on-device inside the osFoundry desktop app: files are indexed in local SQLite with a built-in vector store, so folder contents never have to leave your machine to be searchable.
- Fuses the three rankings with reciprocal-rank fusion instead of picking one method, so 'handle auth' can surface authenticateUser() and a recently edited config in the same result set.
- Incremental indexing watches selected folders and re-indexes only what changed, keeping the index current without rescanning everything on each edit.
- Feeds your chats and agents: retrieved local files become grounded context for any model you run under BYOK, with no reseller markup on the model calls.
Use cases
- Developer: type 'handle auth' and osIndex surfaces the exact AuthToken symbol, a function named checkSession, and the config you edited this morning in one ranked list.
- Researcher: search a folder of notes and PDFs by idea rather than exact wording, then feed the top hits to a model as grounded context without the files leaving your laptop.
- Privacy-conscious professional: keep a folder of client documents fully searchable on-device, so retrieval never uploads the contents and only model calls you explicitly enable touch the network.
Frequently asked questions
Does osIndex send my files to the cloud to search them?
No. osIndex indexes your selected folders into local SQLite with an on-device vector store, and both the index and your queries stay on the machine. Local file search does not require uploading folder contents. Some optional osFoundry features (like cloud model calls for grounding an answer, or workspace-level scoring) are gated behind explicit consent toggles, so anything leaving the device is something you turn on deliberately.
What do the three scorers actually do differently?
The lexical scorer matches exact words and symbols, so precise identifiers and phrases rank high. The semantic scorer uses vector embeddings to match meaning, so a query finds relevant files even when they use different wording. The recency scorer weights files you have edited or opened recently. osIndex fuses all three rankings together rather than choosing one, which is why it handles both 'the exact function name' and 'the thing I was working on about billing' well.
How does osIndex stay current as I edit files?
Indexing is incremental. A folder watcher detects changes and re-indexes only the affected files instead of rescanning the whole folder, so the index reflects your latest edits without a heavy full rebuild. New and changed files become searchable shortly after you save them, and recently touched files also get a recency boost in ranking.